Theoretical and real-world applications of superior face recognition.

Theoretical and real-world applications of superior

face recognition

Anna Katarzyna Bobak

A thesis submitted in partial fulfilment of the requirements of

Bournemouth University for the degree of Doctor of Philosophy

Acknowledgements

First and foremost, I would like to thank my supervisor, Dr Sarah Bate, for her on-going support and believing in me when things got tough. I would also like to extend my thanks to Dr Ben Parris and Professor Changhong Liu for their advice throughout my studies. I also offer my sincere appreciation to Bournemouth University for giving me this incredible opportunity to pursue a career as a researcher. My special thanks are extended to all participants -this project would have not been possible without their help and dedication.

My deepest gratitude goes to my Mum and Grandparents- thank you, not only for your unconditional love and support throughout my studies, but also for bringing me up who I am today. I love you. To my cousin Gabi- thank you for all the chats and for being the voice of common sense when I needed it the most.

My warmest thanks go to my friend, Kinga. Thank you for the long chats, visits, and immediate support when things were not looking up, but also for celebrating all the achievements with me. To Magda, Mikael, and Milton, thank you for being there for me and never stopping to believe that I have got what it takes. To my dear friends Evaggelia, Dina, and George, thank you for your unconditional support and for making me feel like a part of your family.

My completion of this thesis would not have been possible without the caring friendships of those up north: Ola, Dagmara, Patryk, Ania, Clare, Ailsa, and Emma- thank you for being there for me along the way. To Jenny, Jamie, and Helen- thank you for keeping me sane at home and being such a positive influence in my life. I would also like to thank Kate and Charlie for their friendship and care. Furthermore, I would like to thank

Becky, Ruth, and Andy for their warm support and for motivating me to swim, bike, and run in the past year.

My friends, Simon, Nabil, Abby, Michele, Jess, Lesley, Becca, Sarah, and Emmy deserve a special mention. You made the PhD journey much more enjoyable and less lonely. I am also extremely grateful to Martin and Jamie for their continuous support with my project.

Last but not least, I would like to thank the late Dr Robbie Cooper for sparking my interest in faces and for inspiring me to become a scientist. This one is for you.

Copyright Statement

This copy of the thesis has been supplied on condition that anyone who consults it is understood to recognise that its copyright rests with its author and due acknowledgement must always be made of the use of any material contained in, or derived from, this thesis.

Abstract

While previous work has identified the existence of people with extraordinary face recognition skills (so-called “super-recognisers”; SRs), the cognitive and perceptual underpinnings of the ability are unknown. This thesis addresses this issue, using behavioural and eye-movement measures. It also evaluates the methods used to identify SRs, their role in more applied national security settings, and ways of improving face recognition in typical perceivers. The first set of studies offers an in-depth cognitive and perceptual examination of six SRs using a case-series approach. This investigation revealed that while SRs are a heterogeneous group, they consistently show enhanced holistic processing. A second set of studies examined the eye-movements of SRs in a standard face memory task and a more ecologically valid free-viewing task. In both experiments SRs spent more time looking at the nose (i.e. the centre of faces) than typical perceivers, countering previous work that suggests the eye region is critical in facial identification. A subsequent study was aimed at establishing the UK-specific norms for dominant tests of face recognition and face perception, using a large sample of young British adults. Results suggested that females are better at face recognition than males, and that country-specific control norms are needed for these neuropsychological tests. A fourth set of studies looked at the performance of SRs on more applied face recognition tasks, replicating face matching and recognition scenarios. Results strongly suggested that some SRs are best-suited to particular tasks, and when identified correctly would make extremely valuable employees in national security settings. A final study examined if face matching and face recognition skills can be improved in typical perceivers via intranasal inhalation of the nonapeptide oxytocin, yet neither process was improved following this intervention. The theoretical and practical implications resulting from all these

investigations are discussed, particularly in relation to our understanding of the typical face-processing system, and in making practical recommendations for the implementation of super recognition in national security settings.

Table of Contents

Acknowledgements

... ii

Copyright Statement

... iv

Abstract

...v

List of Figures

... xi

List of Tables

... xii

Author’s Declaration

... xiv

Chapter 1: Introduction ...1

1. THE FACE RECOGNITION SPECTRUM ... 3

2. THE IDENTIFICATION OF SUPER RECOGNITION ... 6

3. THE COGNITIVE UNDERPINNINGS OF SUPER RECOGNITION ... 8

4. PROCESSING STRATEGIES IN SUPER RECOGNITION ... 10

5. ARE SUPER RECOGNISERS USEFUL IN APPLIED SECURITY SETTINGS? ... 13

6. THESIS OVERVIEW AND AIMS ... 21

Chapter 2: An in-depth cognitive examination of individuals with superior

face recognition skills ... 22

1. INTRODUCTION ... 23

2. CASE DESCRIPTIONS ... 27

3. STUDY 1: IS SUPER RECOGNITION FACE-SPECIFIC? ... 31

3.1. Matching test ... 32

3.2. Object memory ... 37

3.3. Summary of Study 1 ... 38

4. STUDY 2: LOCATING SUPER RECOGNITION WITHIN THEORETICAL MODELS OF FACE-PROCESSING ... 38

4.1. Perception of facial identity ... 39

4.2. Perception of emotional expression ... 41

4.3. Summary of Study 2 ... 43

5. STUDY 3: PROCESSING STRATEGIES IN SUPER-RECOGNITION ... 43

5.1. The Navon task... 44

5.2. Inversion effects ... 44

5.4. Summary of Study 3 ... 51

6. THE COGNITIVE HETEROGENEITY OF SUPER RECOGNITION ... 51

7. GENERAL DISCUSSION ... 54

7.1. Domain-specificity of super recognition ... 54

7.2. Locating superior recognition within cognitive models of face-processing ... 55

7.3. Processing strategies in super recognition ... 56

7.4. The cognitive heterogeneity of super recognition ... 59

7.5. Conclusion ... 61

Chapter 3: Eye-movement strategies in developmental prosopagnosia and

“super” face recognition ... 62

1. INTRODUCTION ... 63 2. EXPERIMENT 1 ... 66 2.1.Method ... 66 2.2. Results ... 73 2.3. Summary of Experiment 1 ... 79 3. EXPERIMENT 2 ... 79 3.1. Method ... 80 3.2. Results ... 81 3.3. Summary of Experiment 2 ... 88 4. EXPERIMENT 3 ... 88 4.1. Method ... 89 4.2. Results ... 90 4.3. Summary ... 93 5. DISCUSSION ... 93

Chapter 4: The identification of super recognition in young British adults .... 100

1. INTRODUCTION ... 101 2. METHOD ... 106 2.1. Participants ... 106 2.2. Materials ... 107 2.3. Procedure ... 109 3. RESULTS ... 109 3.1. CFMT+ ... 109 3.2. CFPT ... 113

3.3. Do young adults have insight into their face recognition ability?... 114 4. GENERAL DISCUSSION ... 115 4.1. CFMT+ ... 116 4.2. CFPT ... 117 4.3. Gender effects ... 118 4.4. Self-report... 121 4.5. Conclusion ... 122

Chapter 5: Applied value of super-recognition: Evidence from line-up and

face recognition paradigms ... 123

1. INTRODUCTION ... 124 2. EXPERIMENT 1 ... 126 2.1. Method ... 126 2.2. Results ... 131 2.3. Discussion ... 136 3. EXPERIMENT 2 ... 136 3.1. Method ... 137 3.2. Results ... 139 3.3. Discussion ... 143 4. GENERAL DISCUSSION ... 144

Chapter 6: Solving the border control problem: Evidence of enhanced face

matching in individuals with extraordinary face recognition skills ... 151

1. INTRODUCTION ... 152

2. METHOD ... 154

3. RESULTS ... 157

Glasgow Face Matching Test ... 157

Models Face Matching Test ... 162

4. GENERAL DISCUSSION ... 166

Chapter 7: Improving face recognition in forensic settings: The effects of

oxytocin on face recognition and face memory tasks ... 172

1. INTRODUCTION ... 173

2. EXPERIMENT 1 ... 176

2.1. Method ... 176

2.3. Summary ... 181 3. EXPERIMENT 2 ... 181 3.1. Method ... 182 3.2. Results ... 184 3.3. Summary ... 186 4. GENERAL DISCUSSION ... 187

Chapter 8: General Discussion ... 193

1. THE NEUROPSYCHOLOGICAL “DIAGNOSIS” OF SUPER RECOGNITION ... 194

2. THEORETICAL PERSPECTIVES ... 200

3. PRACTICAL IMPLICATIONS FOR FORENSIC AND SECURITY SETTINGS ... 204

4. SUMMARY AND FUTURE DIRECTIONS ... 211

References

... 212

List of Figures

Chapter 1 ...1

Figure 1. Bruce & Young’s (1986) sequential model of face recognition ... 10

Chapter 2: An in-depth cognitive examination of individuals with superior

face recognition skills ... 22

Figure 1. The structure of the CFMT+ (Russell et al., 2009) ... 28

Figure 2. Sample stimuli from the object matching task ... 34

Chapter 3 ... 62

Figure 1. Example stimuli from Experiments 1 and 2 ... 71

Figure 2. The percentage dwell time spent by DPs and controls on each region in Experiment 1 ... 76

Figure 3. The percentage dwell time spent by SRs and controls on each region in Experiment 2 ... 84

Chapter 4 ... 100

Figure 1. Distribution of CFMT+ scores. ... 110

Chapter 5 ... 123

Figure 1. An example trial from a face matching array ... 129

Figure 2. D prime on the 1-in-10 task in Experiment 1 plotted against CFMT score ... 130

Figure 3. Example stimuli sfrom study and test in Experiment 2 ... 134

Figure 4. D prime on the face memory task in Experiment 2 plotted against CFMT score143

Chapter 6 ... 151

List of Tables

Chapter 2 ... 22

Table 1. Demographical and background neuropsychological information about the SR participants ... 30

Table 2. Results from the object-processing tasks administered in Study 1…. ... 36

Table 3. Results from the tasks administered in Study 2. ... 40

Table 4. Results from the configural processing tests described in Study 3 ... 46

Table 5. The overall pattern of performance noted for each of the six SR participants. ... 53

Chapter 3: Eye-movement strategies in developmental prosopagnosia and

“super” face recognition ... 62

Table 1. Demographics of DPs ... 68

Table 2. Performance of DPs and controls on each measure in Experiment 1 ... 78

Table 3. Demographical information, CFMT+ scores for the SR participants ... 82

Table 4. Performance of the SRs and controls on each measure in Experiment 2. ... 86

Table 5: Performance of the SRs and controls on the eye-tracking task. ... 92

Chapter 4 ... 100

Table 1. Gender effects on CFMT+ and CFPT scores for all participants. ... 112

Table 2. Proposed cut-off points for SR in young British adults. ... 113

Table 3. The correlations between single items questions assessing self-perceived face recognition ability and the CFMT and CFPT ... 115

Chapter 5 ... 123

Table 1. Demographical information and CFMT+ scores for the SR participants ... 128

Table 2. Mean (SD) accuracy score for SRs and control participants in Experiment 1. ... 132

Table 3. Individual case analyses of sensitivity of SRs in Experiment1 ... 135

Table 4. Performance of SR and control participants in Experiment 2 ... 140

Table 5. Individual case analyses of sensitivity and response bias of SRs in Experiment 2 ... 141

Chapter 6 ... 151

Table 3. Individual case analyses of sensitivity of SRs in GFMT ... 161 Table 4. Group accuracy descriptive statistics in MFMT ... 162 Table 5. Individual case analyses of sensitivity of SRs in MFMT ... 164

Chapter 7: Improving face recognition in forensic settings: The effects of

oxytocin on face recognition and face memory tasks ... 172

Author’s Declaration

I hereby declare that the work presented in this thesis has not been and will not be, submitted in whole or in part to another University for the award of any other degree.

Human faces convey an array of socially salient information such as identity, gender, and emotional state. The ability to extract this information is critical for appropriate social functioning. While most people have similar levels of experience with faces, there are still considerable individual differences in their ability to recognise facial identity (e.g. Bate, Parris, Haslam, & Kay, 2010; Bowles et al. 2009). These differences range from individuals who are remarkably good at face recognition (so-called “super recognisers”, SRs: Russell, Duchaine, & Nakayama, 2009; Bobak, Hancock, & Bate, 2016) to those affected by developmental prosopagnosia (DP). This latter group of people experience severe difficulties in face recognition, in the absence of neurological damage or illness, lower-level visual or intellectual impairments, and concurrent socio-emotional difficulties (Bate & Cook, 2012; Bate et al., 2014a; Jones & Tranel, 2001; Susilo & Duchaine, 2013).

While a considerable amount of research has examined the correlates of face recognition in both the typical population (e.g., Bowles et al., 2009; Wilmer et al., 2010) and those with face recognition deficits (e.g., Barton, 2008; Behrmann, Avidan, Marotta, & Kimchi, 2005; Le Grand et al., 2006), comparatively little work has focused on the upper end of the face recognition spectrum by examining SRs. The term was first coined by Russell et al. (2009) who identified four people with extraordinary face recognition skills. This group of individuals outperformed control participants on tests of face memory, face perception, and familiar face recognition. However, very little subsequent work has been published on super recognition, and it is unknown whether the superior abilities of SRs extend beyond facial identity processing, nor have the underlying mechanisms of super recognition been identified. From a more applied perspective, it is clear that SRs could potentially be invaluable in policing and national security scenarios. Yet, no published work to date has explored this possibility, and it is likely that

developments in our theoretical understanding of super recognition are required before their more applied potential can be realized. This thesis sets out to address these issues.

1. THE FACE RECOGNITION SPECTRUM

There are large individual differences in the ability to recognize (Bowles et al., 2010; Russell et al., 2009) and perceive (Megreya & Bindemann, 2013; Megreya & Burton, 2006) faces, and particular difficulties are associated with the processing of unfamiliar facial stimuli (see Hancock, Bruce, & Burton, 2000 for a review). A large body of evidence suggests that familiar and unfamiliar faces are processed in a qualitatively different manner. For instance, some people with acquired prosopagnosia are impaired at familiar face recognition but can match unfamiliar faces (e.g., Bauer, 1984; Benton & Van Allen, 1972; Bruyer et al., 1983; Tranel, Damasio, & Damasio, 1988), while others can recognise familiar faces, but are unable to match images of unfamiliar faces (e.g., Young, Newcombe, DeHaan, Small, & Hay, 1993). Malone, Morris, Kay and Levin (1982) identified two patients with damage to occipital areas of the brain. Over time familiar face recognition improved in one patient, but he remained impaired at unfamiliar face matching. Conversely, the other patient remained unable to recognise familiar faces, but was able to match faces that were unfamiliar.

Familiar and unfamiliar faces are also associated with different regional sampling. Ellis, Shepherd, and Davies (1979) found that while familiar face recognition was most efficiently achieved via sampling of the internal facial regions (i.e. the eyes, nose and mouth), unfamiliar face recognition was equally as accurate when the internal or the external features were examined. More recently, Megreya and Burton (2006) examined the recognition of upright and inverted familiar and unfamiliar faces. They found a strong

correlation between the matching of familiar and unfamiliar faces, but only when the familiar faces were inverted. These findings suggest that the two types of facial stimuli are processed in a qualitatively different manner.

In addition to the qualitative differences in familiar and unfamiliar face processing, the matching and recognition of unfamiliar faces is much less stable than it is for familiar faces. Viewpoint, expression, and context (for a review see Johnston & Edmonds, 2009) all have a considerably detrimental effect on unfamiliar face recognition, but only marginally affect the recognition of familiar faces. This phenomenon can be explained by recent research suggesting that more stable representations of an unfamiliar face can be established following increased exposure to a variety of images of that person (e.g. Burton, Kramer, Ritchie, & Jenkins, 2016; Ritchie & Burton, In press). These more detailed representations may be immune to the detrimental effect of the changes listed above, whereas more limited representations of unfamiliar faces may largely rely on pictorial information that prevents recognition under novel viewing conditions.

Unfamiliar face recognition also seems to be related to a number of personality factors. For instance, empathy has been found to be positively related to accuracy of face recognition (Bate et al., 2010). Specifically, Bate and colleagues reported that people high in empathy are better at recognising newly learned faces than those who score low on the empathy scale. This may be because the additional information about others’ emotional state aids their encoding of new faces. Nonetheless, it is also possible that the naturally occurring high empathy makes these individuals allocate more attention to faces. Furthermore, Hills, Eaton, and Pake (2016) reported that psychometric schizotypy, a cluster of traits related to difficulties in social situations (e.g. anxiety), is negatively related to face recognition accuracy. In addition, three studies have shown a direct link between general anxiety and face recognition. In the first report (Mueller, Bailiss, & Golstein,

1979), the authors divided participants into low and high anxiety groups and reported that those low in anxiety performed better in a face recognition task. Furthermore, Megreya and Bindemann (2013) demonstrated that neuroticism and anxiety are negatively correlated with face matching ability, but only in female observers. Another study by Davis, McKone, Dennett et al. (2011) investigated the relationship between social and trait anxiety and the Cambridge Face Memory Test (CFMT) and found that poorer performance on the CFMT was correlated with a significant increase in participants’ social, but not trait anxiety. The authors argued that this suggests successful facial recognition is crucial for social interactions. Indeed, low performance on the CFMT may be related to humans’ perceptual learning, where a particular skill is developed with gradual exposure to (and increased exposure with) a stimulus. As such, frequent social interaction would lead to increased expertise in the within-category discrimination of faces, whilst individuals who do not have that expertise in facial recognition due to increased social anxiety and avoidance of social situations are naturally more likely to underperform at a face recognition task. This account can be supported by studies showing that gregariousness is related to individuals’ face recognition ability through exposure (Arnell & Dube, 2015; Li, Tian, Fang et al., 2010) and that, conversely, shy children are less sensitive to cues necessary for face recognition (Brunet, Mondloch, & Schmidt, 2009). On the other hand, it is also possible that those with poor face recognition skills acquire social anxiety due to problems in everyday life caused by the failure in recognising faces of colleagues, family and friends.

While the factors identified in this section have been found to have small effects on face recognition ability, it remains unclear why some people excel at this task. It is possible that they possess all or many of these factors, or that their skills are simply underpinned by enhancements in visuo-cognitive processes alone.

2. THE IDENTIFICATION OF SUPER RECOGNITION

Temporarily placing aside the issue of the underpinnings of super recognition, another fundamental practical issue is concerned with the identification or “diagnosis” of superior face recognition skills – a topic that has received very little attention to date. The two existing SR papers have primarily identified their SR participants using a cut-off of two standard deviations above the control mean on the long form of the Cambridge Face Memory Test (CFMT+; Russell, Chatterjee, & Nakayama, 2012; Russell et al., 2009). The standard form of this test (the CFMT: Duchaine & Nakayama, 2006) is extensively used to examine individual differences in unfamiliar face recognition skills in both typical perceivers (Bowles et al., 2009; Richler, Cheung, Gauthier, 2011; Wilmer et al., 2010) and those suspected to have DP (Bate et al., 2014; Bate et al., 2008; Russell et al., 2012; Russell et al., 2009), whereas the extended version of the test overcomes ceiling effects associated with the earlier version. It is generally well accepted that the standard form of the CFMT has excellent psychometric properties, with particularly high reliability (Bowles et al. 2009; Duchaine & Nakayama, 2006).

Russell and colleagues (2009) also used a second test to identify SR participants: a “before they were famous” test which presents photographs of celebrities that were taken some time before they became famous. Unsurprisingly, the SRs also performed well on this test, but it is very difficult to use a famous face test as a reliable diagnostic indicator. Indeed, the level of exposure to target faces and to similar tests (these often appear on social media) cannot be controlled between participants. Although Russell et al. (2009) reported positive correlations between this test and performance on the CFMT and CFMT+, a sampling error makes interpretation of these findings difficult, if not

impossible. Namely, four of the 29 participants were SRs in Russell et al.’s study - while there are no published reports on the prevalence of super recognition in the general population, it is highly unlikely that such a high proportion of individuals would possess extraordinary face recognition skills. Hence, the top end of the score distribution in Russell et al.’s study is artificially inflated, and the conclusion that the famous face test correlates with the CFMT and CFMT+ should be seen as tentative. In any case, given most people are excellent at familiar face recognition and individual differences in the face recognition skills of typical perceivers are much better documented in tests of unfamiliar face recognition (Bowles et al., 2009; Richler et al., 2011; Wilmer et al., 2010), there is not currently a strong rationale for using famous face tests to identify SRs.

Russell and colleagues (2009) also examined the perception of facial identity (i.e. by presenting images simultaneously for comparison, placing no demands on face memory) in their four SR participants, using the Cambridge Face Perception Test (CFPT; Duchaine et al., 2007). While Russell et al. make the case that their SRs also outperformed control participants on this test, it should be noted that only a group-based comparison was offered as opposed to the single-case analyses that are typically presented in cognitive neuropsychological investigations (e.g. Bate et al., 2008, 2014). However, it is near impossible for individuals to significantly outperform controls on this test using single-case comparisons given the large variation in control performance and the resulting large standard deviation. Nevertheless, it is of note that examination of the raw data (see Figure 5, Russell et al., 2009) indicates that only some SRs performed above the control mean on the CFPT. This data raises the possibility that the superior face recognition skills of SRs are not always associated with superior face perception skills.

3. THE COGNITIVE UNDERPINNINGS OF SUPER RECOGNITION

The possibility that facial identity perception may not always be facilitated in SRs is important from a theoretical perspective. Cognitive neuropsychological investigation of cases of both developmental (e.g. Bate et al., 2009; Eimer, Gosling, & Duchaine, 2012; Garrido, Duchaine, & Nakayama, 2008) and acquired (e.g. Bate et al., 2015; Rezlescu, Pitcher, & Duchaine, 2012) prosopagnosia have aided the development of cognitive theories of typical face-processing, and tested their assumptions. For instance, the dominant model posited by Bruce and Young (1986) suggests that face-processing is a hierarchical sequential process (see Figure 1), whereby an initial stage of visual analysis is proceeded by the structural encoding of an incoming facial representation. At this stage, the view-dependent representation of the image is transformed into a view-independent representation, in preparation for identity recognition. Once the view-independent representation is constructed, it is compared to all stored representations of known faces in the face recognition units (FRUs). If a familiarity match is achieved, the relevant person identity node (PIN) is activated, and biographical information about that person is retrieved. Finally, the name of the person is accessed. Meanwhile, other perceptual aspects of the view-dependent representation (e.g. emotional expression) are thought to be processed independently to identity recognition.

Investigations using individuals with prosopagnosia suggest that face-processing can be interrupted at different stages of Bruce and Young’s model, and that these patterns of impairment may relate to different subtypes of the condition. These findings broadly relate to deficits in face perception that are thought to occur at the level of structural encoding (e.g. Eimer, 2000; Eimer & McCarthy, 1999), and higher-order deficits affecting only face recognition that have been linked to impairments at the level of the FRUs or the PINs (Rezlescu et al., 2012). This latter group of individuals have also contributed to a key

double dissociation that has bolstered the theory that facial identity and facial expression are processed independently. That is, neuropsychological patients with prosopagnosia alongside intact facial expression recognition skills appear to present with the reverse pattern of impairment to individuals who cannot recognise facial expression but can recognise facial identity.

Cognitive theories of face-processing can therefore be adopted to predict potential patterns of performance in SRs. Specifically, super recognition (a) may be underpinned by enhanced processing at dissociable stages of the face-processing model, and (b) if the enhancement is at the level of structural encoding, the processing of other aspects of face perception (e.g. facial expression) may also be heightened. Conversely, if the enhancement is at the latter stages of processing (i.e. at the level of the FRUs or PINs) only the recognition of facial identity will be facilitated. While the consistency of the cognitive presentation of SRs provides a novel means to test the predictions of theoretical models of face-processing, no in-depth study has addressed these issues to date.

Figure 1. Adaptation of the Bruce & Young’s (1986) sequential model of face recognition.

4. PROCESSING STRATEGIES IN SUPER RECOGNITION

While superior face recognition skills may be underpinned by one or more facilitation within the theoretical framework offered by Bruce and Young (1986), another possibility is that SRs use more efficient processing strategies to extract facial information. This may be reflected in their use of the “configural” or “holistic” processing strategy that is thought to be optimal for successful face recognition (e.g. Richler, Cheung, & Gauthier, 2011), or it may be that they are drawn towards specific regions of the face that hold information that is critical for identification.

Configural and holistic processing

Numerous reports indicate that faces are processed in a different manner to objects (McKone & Robbins, 2011; Rossion, 2013). For example, faces are thought to be processed holistically – that is, information is thought to be integrated from across the face rather than being broken down into individual parts (Piepers & Robbins, 2012; Rossion, 2013; see Maurer, Le Grand, & Mondloch, 2002; Richler, Palmeri, & Gauthier, 2012, for a review of different meanings of “holistic processing”). Furthermore, evidence suggests that we are highly sensitive to relational or configural information in faces (Maurer et al., 2002; Piepers & Robbins, 2012): this means that we are more sensitive to subtle variations in the spacing of facial features compared to those of other objects (e.g. Robbins, Shergill, Maurer, & Lewis, 2011; Yovel & Kanwisher, 2008). There is a long-standing belief that the use of these holistic and/or configural processing styles may underlie our proficiency in face recognition (e.g., Richler, Cheung, & Gauthier, 2011a; Rossion, 2013), and many studies attempting to explain group or individual differences in face processing have examined indicators of configural or holistic processing (e.g., DPs and controls: DeGutis, Cohan, Mercado, Wilmer, & Nakayama, 2012; Palermo et al., 2011; children and adults: Crookes & McKone, 2009; Mondloch, Le Grand, & Maurer, 2002; individual differences: DeGutis, Wilmer, Mercado, & Cohan, 2013; Richler et al., 2011a). Given the apparent importance of holistic and configural information in face recognition, it is possible that super recognition is underpinned by proficiencies in these face-specific perceptual processes.

Some preliminary evidence supports this hypothesis. The SRs reported by Russell et al. (2009) showed a larger face inversion effect (a difference in performance between upright and inverted faces) than control participants. The inversion effect is thought to reflect the fact that face-specific perceptual processes – namely, holistic and configural

processing – are specialised for upright faces, and are disturbed or reduced in inverted faces (Maurer et al., 2002; Richler, Mack, Palmeri, & Gauthier, 2011b). Therefore, a larger inversion effect is thought to reflect stronger holistic or configural processing, and the fact that SRs showed superior performance for upright faces but relatively normal performance for inverted faces indicates that they may be particularly good at these tasks. Further, Sekiguchi (2011) examined eye-movements in a group of participants with typical face recognition abilities, and found that individuals with higher face memory (who were not SRs) tended to make more saccades between the two eyes than participants with poorer face memory (who did not have prosopagnosia). Sekiguchi suggested that this finding may also reflect the importance of configural processing (i.e. processing the specific spatial relationships between the eyes and other facial features) in face recognition.

The use of regional facial information

The finding reported by Sekiguchi (2011) is of interest because it suggests that information from the eye region may be critical for optimal facial identification performance. Many other studies have reported that the eye region is particularly pivotal for the recognition of facial identity, given typical perceivers fixate on the eyes to a greater extent than any other facial region (e.g. Schyns, Bonnar & Gosselin, 2002; Slessor, Riby & Finnerty, 2013). These findings are bolstered by reports that individuals with acquired prosopagnosia spend less time examining the inner features of the face (i.e. the eyes, nose and mouth) than controls, (e.g. Caldara, Schyns, Mayer, Smith, Gosselin, & Rossion, 2005; Lê, Raufaste, & Demonet, 2003; Lê, Raufaste, Roussel, Puel, & Demonet, 2003; Stephan & Caine, 2009), and one study has reported the same effect in DP (Schwarzer et al., 2007). More specifically, some studies suggest that participants with acquired

prosopagnosia (Caldara et al., 2005; Bate et al., 2015; Stephan & Caine, 2009; Van Belle, Ramon, Lefevre, & Rossion, 2010) spend less time examining the eyes and more time examining the mouth than control participants.

However, other lines of evidence suggest the critical measure is not the proportion of dwell time spent on the eyes, but the time spent examining the nose. Hsiao and Cottrell (2008) reported that the optimal viewing position in face recognition (i.e. the location of the first fixation that a person makes to a face) is to the left of the centre of the nose. In addition, the preferred landing position (i.e. the location that participants fixate the most) is around the centre of the nose, rather than within the eye region. Similarly, Peterson and Eckstein (2012) found that the optimal viewing position on a range of face-processing tasks was below the eyes and towards the left side of the nose – in a remarkably similar position to that observed by Hsiao and Cottrell. Both sets of authors suggest that this viewing position may be the optimal location for holistic processing of the entire face to occur.

No study to date has examined eye-movement strategies in SR participants. Investigation of this issue does not only have the potential to inform us about the underpinnings of superior face recognition, but may also inform the key theoretical debate regarding the importance of the eyes versus the nose in facial identification.

5. ARE SUPER RECOGNISERS USEFUL IN APPLIED SECURITY SETTINGS? The above review clearly indicates that investigation of super recognition is an innovative means to garner further insight into our theoretical understanding of the typical face recognition system. However, it is also of practical interest to examine whether SRs

may play an important role in policing and security settings, given excellent face recognition skills are pivotal in some occupations. This is particularly pertinent given increasing findings that (a) typical perceivers (and even those with many years of experience in relevant roles) tend to make many errors in applied face-processing tasks, and (b) the limited success that has been observed in studies that have attempted to improve face recognition skills in the typical population.

The performance of typical perceivers on applied tasks of unfamiliar face-processing

Numerous studies have illustrated the difficulties that typical perceivers encounter in their recognition of unfamiliar faces, and many of these studies have adopted paradigms that mirror face recognition scenarios in security or policing settings. Photographic ID is the most ubiquitous means of evaluating one’s identity. It is commonly used to buy age restricted items, travel, to access one’s work place or to verify identity by the law enforcement agencies. Overall there is a considerable reliance on this method, despite consistent demonstrations that the matching of unfamiliar faces is a very difficult task. Specifically, in the Glasgow Face Matching Test (GFMT, Burton, White, McNeill, 2010), where participants have to compare two simultaneously presented images and decide whether they show the same person or not, average accuracy is 80% on the short and 89% on the long version of this task. This relatively high error rate occurs even though both images from “match” trials were taken on the same day from a frontal viewpoint, and no time restriction is imposed on the participant. Importantly, the range of performance on the long version of this test was reported to be between 62% and 100%, suggesting large individual differences within the general population.

In another study, Bindeman, Aveytisan and Rakow (2012) administered the GFMT to participants over three (Experiment 1) and five (Experiment 2) consecutive days. The authors reported overall accuracy on a par with the normative data provided by Burton and colleagues (2012), but also highlighted that while some individuals performed consistently over successive attempts, the accuracy of others was subject to considerable fluctuations. Interestingly, when the same set of face pairs was presented on each of the three days (Experiment 1), participants’ judgements were not stable in that the same image pairs were often classified differently on each day.

Another occupational setting where unfamiliar face recognition ability is pivotal involves work with video footage or the recognition of missing or wanted persons by officers on patrol. Studies have consistently shown that while such tasks are easy to perform with personally familiar faces (Bruce et al., 2001; Burton et al., 1999), the accuracy of identification for previously studied unfamiliar faces is close to chance in typical observers (Burton et al., 1999). In a seminal study, Burton and colleagues (1999) asked participants to encode ten identities from video footage and later to identify them from photographs. The identities used in the study were of Psychology lecturing staff and the experimental groups included Psychology students, students from non-Psychology courses and police officers. The group who were personally familiar with the identities presented in the video footage (i.e. the Psychology students) were highly accurate in making familiarity judgements for the studied identities, while participants in the unfamiliar groups (students and police staff) were significantly less accurate in their judgements. Interestingly, there were no differences in performance between those who were untrained and inexperienced in face recognition assignments (i.e. students) and police officers who presumably should be acquainted with this type of task.

As discussed above, one possible reason for these low identification rates is that merely pictorial encoding occurs when only one view of the face is available. Specifically, Bruce and Young’s (1986) modular sequential model of face recognition posits that faces are initially encoded using pictorial view dependent representations, but with gradual exposure to many views of the same face, this view-dependent information is transformed into more stable view-independent representations. This theoretical account is supported by a plethora of empirical evidence suggesting that variability is critical in the learning of new faces (Andrews, Jenkins, Cursiter, & Burton, 2015; Burton, Kramer, Ritchie, & Jenkins, 2016; Dowsett, Sandford, & Burton, 2016; Ritchie & Burton, In press).

These findings clearly suggest that familiarity, likely achieved by the building of stable face representations, increases the likelihood of a positive identification of a face. However, in real life, perpetrators of crimes or missing persons are often unfamiliar to those searching for them, and the number of images available for comparison is restricted by the availability of CCTV footage, the number of images in a database, or the number and quality of photographs provided by family and friends of a missing person.

The considerable individual differences in face matching performance have important implications for work assignments in national security settings. Pertinently, passport controllers, victim identification officers and CCTV operators perform facial image comparisons on an everyday basis. What is more, routine ID checks accompanying the purchase of age-restricted items are performed by individuals drawn from the typical population. Pertinently, while the images used in the aforementioned studies of individual differences can be described as “optimal” for face matching to be performed accurately, in real life one’s appearance is subject to changes in age, weight, hair style, lighting, and so on. Indeed, in a study by Megreya, Sandford, and Burton (2013), participants’

performance on a version of the GFMT (Experiment 2) was 90% when images were taken on the same day, but declined to 70% when images were taken several months apart.

In more realistic settings, Kemp, Towell, and Pike (1997) examined fraud detection from credit cards in a supermarket setting. Cashiers were assigned to a task of matching photographic credit cards (with images sized 2cm x 2cm) to their bearers at the checkout. When cards were used falsely and the identity of the shopper did not match the credit card, accuracy amongst experienced cashiers was as low as 50%. This was despite the fact that the supermarket employees were briefed before the study and promised a bonus of 50% above their statuary reimbursement if they performed the checks accurately and in a timely fashion. As such, participants in Kemp et al.’s study were presumably motivated to be vigilant and perform well. It is possible that in real-life situations and in the absence of these financial inducements accuracy would be even lower.

When IDs are used for international border crossing, successful matching of photographic documents to their holders is a matter of national security. In a recent study White, Kemp, Matheson, Jenkins, and Burton (2014) showed that when compared to unqualified students, experienced passport control officers make a comparable amount of errors when fraudulent mock IDs are produced. Officers were reported to make 10% of mistakes, even though (1) the photographs for mock IDs were taken only a few days prior to the testing session and (2) the foils (photographs depicting a person different to the ID bearer) were chosen opportunistically from students who volunteered to take part in the study, rather than being carefully chosen according to pre-rated similarity to the holder. It is conceivable that if the photographs were taken many months apart, the accuracy would decline similarly to performance in the paradigms used by Megreya and colleagues (2013).

Together, the above literature illustrates that, across many tasks, unfamiliar face recognition is a difficult and error-prone process. It is therefore conceivable that identification of SRs as potential candidates for some real-word occupations would be invaluable for national security. Whether the existing means of identifying SRs are useful for more applied face recognition scenarios is as of yet unknown, and of course if theoretical investigations do uncover different subtypes of super recognition, the same individuals may not necessarily be suited to every task.

Can face recognition skills be improved?

In order for SRs to be used for more applied roles they must be readily available. The precise definition and prevalence of “true” SRs has not yet been determined, raising the possibility that there may not be many of these individuals available for employment. An alternative is that typical perceivers may be trained to become SRs, or at least to undergo some improvement in their face recognition skills. However, the study described above by White et al. (2014) casts doubt on whether this is possible, given they found no difference in face recognition performance according to officers’ years of relevant experience. Three other studies have reported that trained experts excel at face matching tasks in comparison to typical perceivers (Norell et al., 2014; White, Dunn, Schmid,, & Kemp, 2015a; White, Phillips, Hahn, Hill, O’Toole, 2015b). However, baseline face recognition ability was not examined in those participants and it is unclear whether they have a natural ability to process unfamiliar faces or if their superior performance is a result of training and experience. It is thus possible that these “experts” are aware of their extraordinary face recognition ability and self-select for assignments involving facial image comparison. It is also important to note that neither of the recent studies with

experts reported case-by-case analyses or the variation in performance within the experimental groups. As such, these studies shed little light on the processes underpinning superior face matching in the expert groups.

While the error-prone performance in various face matching paradigms has been well-documented in the literature, only recently have efforts been made to enhance it, mostly with limited success. Two investigations concentrating on facial features (Woodhead, Baddeley, & Simmonds, 1979) and face shape (Towler, White, Kemp, 2014) were not able to improve face matching performance at all. Attempts to improve face matching with trial-to-trial feedback have yielded mixed results. While White, Kemp, Jenkins and Burton (2014) reported improvement in face matching performance generalising to novel faces, Alenezi and Bindemann (2013) found that trial-to-trial performance feedback merely inhibits performance decline, but does not lead to an overall increase in matching accuracy. Moore and Johnson (2013) investigated the impact of food incentive, on performance in a face matching task. They found that the overall discriminability increased when participants were made aware of a sweet food incentive, and participants also became more conservative in their responding, i.e. the improvement was driven by the increased accuracy on mismatched trials. Other attempts at improving face matching performance included caricaturing (McIntyre, Kittler, Hancock, & Langton, 2012) and redesigning the ID to include multiple photographs of the holder (White, Burton, Jenkins, & Kemp, 2014), all yielding rather limited results.

A recent study by Dowsett and Burton (2015) examined the impact of working in pairs on face matching performance. In a series of experiments, the authors showed consistent increase in face matching accuracy and further individual improvement in performance, particularly in those whose scores were initially low. Most importantly, the effect of working in pairs was transferable to individual performance on a new set of

images, a finding providing a potential path for future regimes. The longevity of this effect is, however, unclear as all participants were tested on the same day. Furthermore, in the current economic climate and governmental agencies affected by austerity measures, increase in staffing is an unlikely step. As such, ways of improving individual performance are a pivotal research avenue.

One such attempt was recently reported by Bate et al. (2014) where participants inhaled a nasal spray with oxytocin or placebo before completing the one-in-ten task (Bruce et al., 1999), a well-established line-up matching paradigm containing target- present and target-absent trials. While, participants in the oxytocin condition had better accuracy on target-present trials, they were also more prone to make positive identifications in target-absent trials, a potentially costly mistake, when made within the national security or forensic sectors. Bate et al. (2014) suggested that it is possible that the aspects that contribute to oxytocin’s facilitative effect when administered before encoding in face memory tasks, such as increasing the saliency of studied faces, are disadvantageous in line-up scenarios. In target absent line-ups, when many faces are presented simultaneously, participants could be mistaking saliency for familiarity and choosing a foil that seems as the most similar to the target. This is in contrast to the placebo condition, where without the saliency enhancing effect of oxytocin, participants responded in a more conservative way. This latter methodology therefore may still have potential for improved face recognition when the hormone is inhaled before recall, yet this possibility had not yet been explored.

6. THESIS OVERVIEW AND AIMS

In sum, four key questions are addressed in this thesis. First, what are the underpinnings of super-recognition? Second, how prevalent is super-recognition, as measured by the currently available tests of face recognition (the CFMT+, Russell et al., 2009) and face perception (CFPT, Duchaine et al., 2007)? Third, how well do super-recognisers perform on tests of face recognition and face matching resembling real-life scenarios? Finally, is it possible to instantaneously improve face recognition and face matching ability in typical observers?

To answer these questions, studies in Chapter 2 offer an in-depth examination of the cognitive and perceptual abilities of SRs. In addition, eye-movement patterns are examined in Chapter 3 and compared to those of typical perceivers and individuals with developmental prosopagnosia. Chapter 4 investigates the distribution of face recognition ability and scrutinises the currently used cut-offs for the “diagnosis” of super recognition. Chapters 5 and 6 examine the performance of SRs on applied tasks of face matching and memory. Specifically, Chapter 5 looks at the recognition of faces from CCTV and matching performance in the well-established one-in-ten task (Bruce et al., 1999). Chapter 6 further examines the ability of super-recognisers to match faces in tasks resembling passport control and other types of facial image comparison. Chapter 7 investigates the ostensible role of oxytocin in face recognition ability by applying it to the real-life scenarios investigated in the previous two chapters. The final chapter pulls together all these findings and discusses their implications for the diagnosis of superior face recognition, our theoretical understanding of the face-processing system, and practical

Chapter 2: An in-depth cognitive examination of individuals with

superior face recognition skills

Chapter (except for Study 2) published as:

Bobak, A. K., Bennetts, R. J., Parris, B. A., Jansari, A., & Bate, S. (In press).

An In-depth Cognitive Examination of Individuals with Superior Face

1. INTRODUCTION

Russell, Duchaine, and Nakayama (2009) presented the first report of super recognition, describing four individuals who outperformed control participants on tests of face memory, face perception, and familiar face recognition. Specifically, Russell and colleagues reported the face memory and face perception scores of these super-recognisers (SRs) and found that they significantly outperformed typical observers on both types of tasks (in group-level analyses). Given only one other paper to date has investigated super recognition (and that paper focused on the role of surface and reflectance processing in super-recognition; Russell, Chattejee, & Nakayama, 2012), there has been very little work examining the skills that underpin super recognition. In particular, it is unclear whether it is underpinned by enhancements to more generalized mechanisms, specific stages of the face recognition process, or specific processing strategies. Investigation into these issues is of clear theoretical importance, particularly as SRs (akin to those with prosopagnosia) may not represent a homogenous group of individuals in terms of their cognitive presentation or the mechanisms underpinning their superior skills.

This chapter extends the existing SR literature by reporting a detailed

neuropsychological assessment of six individuals who meet the previously published criteria for super recognition. Four questions are addressed via a battery of

neuropsychological and cognitive tests. First, more general perceptual and cognitive processing mechanisms are examined, to investigate whether enhancements in these processes support superior face recognition skills. Much research supports the hypothesis that face recognition is a highly specialised process involving a number of dedicated neural circuits (Haxby, Hoffman, & Gobbini, 2000; Gobbini & Haxby, 2007), and this theoretical standpoint is supported by findings that some individuals with developmental (Duchaine & Nakayama, 2005; Jones & Tranel, 2001) and acquired (Busigny, Joubert, Felician,

Ceccaldi, & Rossion, 2010; de Renzi & di Pellegrino, 1998) prosopagnosia only have difficulties in the recognition of faces. Further, existing work has failed to find a

relationship between face recognition skills in the typical population and performance on tests of non-facial visual memory (e.g. an abstract art memory test) or verbal memory (e.g. verbal paired-associates test) (Wilmer et al., 2012; Wilmer et al., 2010). However, no work to date has examined the domain-specificity of super recognition, and it is possible that particularly good general perceptual or mnemonic abilities could support the exceptional face recognition skills observed in these individuals. Alternatively, if it is found that the exceptional skills of SRs are restricted only to the processing of faces, this would further support the face-specificity hypothesis.

Second, we investigate whether SRs are only proficient at facial identity recognition, or whether their skills extend to other aspects of face-processing (e.g. the recognition of emotional expression). This speaks to important theoretical questions concerning the structure and function of the face-processing system, given dominant cognitive theories posit that the face-processing pathway is composed of a set of

hierarchical sub-processes (e.g. Bruce & Young, 1986). Individuals at the other end of the face recognition spectrum, i.e. those with prosopagnosia (Bate & Bennetts, 2014; Bate & Cook, 2012; Bennetts, Butcher, Lander, & Udale, 2015), can present with or without impairments in face perception (Dalrymple, Garrido, & Duchaine, 2014). These findings have aided the development of dominant models of face-processing by indicating that the face-processing pathway can be lesioned at different locations (i.e. at an early stage involving structural encoding or a later stage involving retrieval); yet the presumed hierarchical nature of the framework explains why the hallmark deficit in facial identity recognition presents even in the former group of individuals. Likewise, it follows that super recognition may result from relatively early enhancements affecting all aspects of

face perception (i.e. judgments that use view-dependent representations, such as age, gender or expression), facial identity perception alone (i.e. after view-independent

representations have been created), or from later enhancements affecting only memory for faces. Examination of the face perception abilities of SRs will therefore have important theoretical implications by presenting a novel means to evaluate current theoretical frameworks.

Third, the processing strategies used by SRs are examined, to investigate whether these are different or merely enhanced in comparison to typical perceivers. Indeed, while SRs are clearly better than controls at face recognition in quantitative terms, it is unknown whether this heightened performance is underpinned by qualitative differences in

processing strategy, whereby they use different types of visual information or different information processing styles. Alternatively, SRs may adopt the same processing strategies as typical perceivers, but in a heightened or more efficient manner. It is well-accepted that configural or holistic processing strategies underpin typical face recognition and many researchers agree that the composite task is the most robust measure of this process in group studies (Richler, Floyd, & Gauthier, 2014). Although this task has not yet been administered to SRs, existing work indicates that it does correlate with face recognition abilities in the general population (Richler et al., 2011; Wang, Li, Fang, Tian, Liu, 2012, c.f. Konar, Bennett, & Sekuler, 2010), and that the composite face effect is reduced in people with prosopagnosia (Avidan, Tanzer, & Behrman, 2011; Palermo et al., 2011; but see Susilo et al., 2010). To date though, no studies have addressed this question directly in SRs, and it remains unclear whether these face-specific processes underpin superior face recognition skills. It should be noted, though, that performance on composite tasks widely varies even in the typical population, and it can be very difficult to reliably detect

Hence, while any significant individual differences using case-by-case analyses may be insightful, null effects are more inconclusive.

It is important to note that holistic processing also occurs on a more general scale (e.g., integrating many different objects into a coherent visual scene). Manipulating this general process by asking individuals to focus on local details (e.g. the small letters in a Navon stimulus) can be detrimental to face recognition, possibly because it encourages piecemeal, non-holistic processing (e.g. Gao, Flevaris, Robertson, & Bentin, 2011; Macrae & Lewis, 2002). Building on this work, some research into prosopagnosia has established that some people with face recognition deficits show a general bias towards the processing of local details, and this correlates with their reduced holistic processing of faces (Avidan et al., 2011; but see Duchaine, Yovel, & Nakayama, 2007). This work suggests that it may be variation in this more general holistic processing ability, rather than a face-specific process per se, that underpins individual differences face recognition abilities. Once again, though, this issue has not been addressed in the SR population. Examination of face-specific configural and holistic processing, alongside more general holistic processing tendencies (also sometimes referred to as “global processing biases”, e.g., Behrmann et al., 2005; Duchaine et al., 2007), would therefore provide insight into whether SRs show specific qualitative differences in processing strategies, and whether these effects are domain-specific or reflect more general perceptual processes.

Finally, we pull our findings together to examine whether SRs show a consistent pattern of enhanced abilities, or whether these individuals vary in their cognitive presentation as has been observed at the bottom end of the face-processing spectrum (i.e. in DP).

2. CASE DESCRIPTIONS

Following widespread media coverage about super recognition, the six individuals described in this chapter contacted our laboratory. DF is an 18 year-old right-handed male Engineering student, TP is a 35 old right-handed male IT manager , GK is a 33 year-old right-handed male university lecturer, JN is a 35 year-year-old right-handed female sourcing consultant, CH is a 27 year-old right-handed male lawyer, and CW is a 21 year-old Psychology graduate.

In an initial informal interview, all the SRs described extraordinary face recognition skills that had been present from an early age. They reported that they are able to recognise people even after a brief encounter or after many years have passed (for instance, childhood friends): “I recently saw a girl who I taught for a couple of swimming lessons when I was a teenager. I recognised her immediately, despite the fact that I had not seen her since she was 6, and she is now 18” (CH). Following existing procedure, each participant was screened using the CFMT+ (Russell et al., 2009). The initial three stages of this test replicate the standard version of the CFMT (Duchaine & Nakayama, 2006). In the encoding stage, participants view each of six novel faces from three different viewpoints, and complete three test trials per face where they are asked to select the encoded identity from a triad of faces. In the second stage, participants are asked to select the encoded identities from novel viewpoints or lighting conditions over 30 triads. The third stage is similar to the second, but the test faces are overlaid with visual noise to make recognition more difficult (24 trials). The CFMT+ then adds a fourth, more challenging stage to the original test: participants are asked to identify the learnt faces from profile images which now display hair, are heavily cropped, or show a different emotional expression (30 trials, see Figure 1).

All six SRs achieved CFMT+ scores that are above the previously-used cut-off of 90/102 (Russell et al., 2009, 2012) on this test (see Table 1). However, we also collected our own control data (N = 30, 15 female; mean age = 25.9 years, SD = 4.5) to ensure that we were comparing our SRs to an appropriately matched control group. Single case statistics showed that all the SRs but one (TP) significantly outperformed the control group: CW and GK, t(32) = 2.66, p = .01, Zcc = 2.70, 95% CI [1.917, 3.474];

estimated % population below their scores = 99.37 and JN, CH and DF, t(32) = 2.40, p = .02, Zcc = 2.445 (95% CI: 1.718 – 3.160); estimated % population below their scores =

98.86. Given TP reached the criteria for super recognition based on previously published control data (Russell et al., 2009) and on additional tests of face recognition (see Chapter 5), we still included him in our sample for this investigation.

All SRs reported normal or corrected-to-normal vision. General intelligence was assessed using the Wechsler Abbreviated Scale of Intelligence, Second Edition (WASI-II, Wechsler, 2011). One SR performed within the “average” range (JN), whereas TP, DF, CH and GK were within the “superior” range (see Table 1). Due to limited time availability, CW’s intelligence was estimated using the WTAR (Holdnack, 2001). Similarly to JN, he scored within the “average” range. While CH excelled at the verbal component of the measurement, DF and JN showed a clear advantage on the performance rather than verbal sub-tests. Conversely, both TP and GK performed similarly on the two sub-tests. This variation in IQ is in line with findings that face recognition ability is domain-specific and unrelated to general intelligence (Wilmer et al., 2009; Zhu et al., 2010).

For each of the investigations below, performance of the SRs was compared to controls using at least two tests to address each theoretical question. For each individual test, a subset of individuals were extracted from a control group containing 30 gender- and age-matched participants (19 female, M age = 32.1, SD = 9.3; see Table 1).

Table 1. Demographical and background neuropsychological information about the SR participants, presented in comparison to controls. Values

for the performance of the SR participants on the CFMT+ are expressed in the number of SDs away from the control mean.

Controls Super-Recognisers Mean SD N CH DF JN GK CW TP Age 32.1 9.3 30 27 18 35 33 21 35 Gender 19 (F) - 30 M M F M M M Handedness 3L - 30 R R R R R R WASI-IIa: Verbal - - - 148 114 99 118 - 127 Performance - - - 111 131 116 119 - 127 Full-2 IQ - - - 134 125 108 121 - 130 WTARb 113.8 8.2 30 - - - - 115 - CFMT+c 68.4/102 11.7 30 2.4* 2.4* 2.4* 2.7* 2.7* 2.0

* indicates participant significantly differed to controls using Crawford et al.’s (2010) modified t-tests for single-case comparisons ( p < .05)

a_{Wechsler Abbreviated Scale of Intelligence, Second Edition (Wechsler, 2011) – this more thorough assessment of IQ was carried out with}

available SRs; bWechsler Test of Adult Reading (Wechsler, 2001) – this quick IQ screen was used with controls to ensure they were

appropriately matched to the SRs and with CW due to time constraints; cCambridge Face Memory Test - Long Form (Russell et al., 2009) – this test was used to confirm superior face recognition skills in the SRs and typical skills in the controls

These individuals had typical face recognition skills (as confirmed by their performance on the CFMT+: see Table 1). Note that a larger control sample is reported for the CFPT, due to the larger variability in the typical population on this test (see below). All control participants presented with normal visual acuity and contrast sensitivity. Not all control participants completed all tests due to time constraints and some computer errors (the N for individual tests is presented in Tables 1-4; gender was approximately equal for each test). For each test, the SRs were compared to the controls on a single case level, using modified t-tests for single case comparisons (SINGLIMS, Crawford, Garthwaite, & Porter, 2010) or Revised Standardised Differences Tests (RSDT, Crawford et al., 2010) as appropriate. This is a particular strength of this work as previous studies (Russell et al. 2009; Russell et al., 2012) have only used group-based statistics to analyse the performance of a smaller number of SRs. Informed consent was obtained from all participants, and ethical approval for the study was granted by the departmental ethics committee.

3. STUDY 1: IS SUPER RECOGNITION FACE-SPECIFIC?

As discussed above, previous work examining super recognition has focused exclusively on their face recognition performance, and it remains possible that the skill is supported by enhancements in more generalized cognitive, perceptual or mnemonic skills. An initial investigation sought to address this issue by examining performance on two different object-processing tests: one assessing matching skills, and the other memory skills.

3.1. Matching test

An object and face matching test was created to assess whether SRs show superior object processing skills compared to typical participants. Participants completed a sequential same/different matching task with faces, hands, and houses (see Figure 2). Each trial consisted of two sequentially presented objects – the initial study image was displayed for 250 ms, and the second test image was displayed until the participant responded. In the face condition, the study image showed a face from a frontal viewpoint and the test image showed a face from a 30-45° angle. Faces were drawn from the Cambridge Face Memory Test–Australian (McKone et al., 2011) and the Bosphorous Face Database (Savran et al., 2012), and were edited to remove external features. Houses were created using the software Realtime Landscaping Plus (Idea Spectrum Inc., 2012). Each house contained the same number of features (three sets of windows and a door), placed onto a constant background texture. The shape and location of the features, the luminance of the background texture, and the overall shape of the house varied throughout the set. As in the face condition, the study and test images presented the houses from two different viewpoints (frontal and 15° profile). Hand images were extracted from the Bosphorus Hand Database (Dutağacı, Yörük & Sankur, 2008), and showed the palm and fingers of a hand. Images were chosen to exclude rings, watches, cuffs, or other identifying features. Study and test images showed the hands in two different positions (e.g., fingers splayed and fingers together), with the wrist pointing downwards (upright condition) or upwards (inverted condition). Each category contained 32 pairs of images (16 same identities, 16 different identities). All pairs were presented twice upright and twice inverted. Trials were blocked by stimulus type, with upright and inverted trials presented randomly within each stimulus type. The order of blocks was randomised between participants. The measure d’

(a bias-free measure of sensitivity; MacMillan & Creelman, 2005) was calculated for